Adapting to Fragmented and Evolving Data: A Fisher Information Perspective

TL;DR

FADE is a Fisher information-based framework for online adaptation to evolving data distributions in machine learning, achieving improved robustness without requiring task labels or replay, and applicable to federated learning.

Contribution

Introduces FADE, a novel Fisher information geometry-based method for robust online learning under sequential covariate shift, with theoretical guarantees and broad applicability.

Findings

Up to 19% accuracy improvement under severe shifts.

Outperforms prior methods like TENT and DIW across benchmarks.

Effective in federated learning with heterogeneous clients.

Abstract

Modern machine learning systems operating in dynamic environments often face \textit{sequential covariate shift} (SCS), where input distributions evolve over time while the conditional distribution remains stable. We introduce FADE (Fisher-based Adaptation to Dynamic Environments), a lightweight and theoretically grounded framework for robust learning under SCS. FADE employs a shift-aware regularization mechanism anchored in Fisher information geometry, guiding adaptation by modulating parameter updates based on sensitivity and stability. To detect significant distribution changes, we propose a Cramer-Rao-informed shift signal that integrates KL divergence with temporal Fisher dynamics. Unlike prior methods requiring task boundaries, target supervision, or experience replay, FADE operates online with fixed memory and no access to target labels. Evaluated on seven benchmarks spanning vision, language, and tabular data, FADE achieves up to 19\% higher accuracy under severe shifts, outperforming methods such as TENT and DIW. FADE also generalizes naturally to federated learning by treating heterogeneous clients as temporally fragmented environments, enabling scalable and stable adaptation in decentralized settings. Theoretical analysis guarantees bounded regret and parameter consistency, while empirical results demonstrate FADE's robustness across modalities and shift intensities.